Methodology

Data

No page on these sites calls a third-party API while you are reading it. Every dataset is fetched on its own cadence by a scheduled job, stored locally, and served from that store. That makes pages fast and — more importantly — means a source going down changes nothing a reader sees except the "as of" date.

Every dataset, its source, its licence, its update cadence and its full contents are published at /data, as JSON and CSV. Provenance is stored with the data rather than typed into a template, so the source line under a chart is generated from the same record that fetched it and cannot drift out of date.

Missing and stale data

A dataset is marked stale once it has gone longer than twice its own cadence without a successful fetch — so a daily source is stale after two days, an annual one after two years. A stale dataset still renders: the chart shows the last good data with a visible "last updated" badge rather than an error or an empty box.

Gaps within a series are drawn as gaps. Lines break across missing values rather than bridging them, because a straight line across a gap is a claim about data that does not exist.

Currently behind cadence:

• Layer 7 attack share — Not yet fetched
• Bilateral trade concentration — Not yet fetched
• EU consolidated sanctions designations — Not yet fetched
• Policy and market rates — Not yet fetched
• Conflict and protest events — Not yet fetched
• Conflict events and fatalities — Not yet fetched
• Event tone by country — Last updated 13 Sep 2026
• Forward election calendar — Not yet fetched
• Liberal democracy index — Not yet fetched
• European electricity load and flows — Not yet fetched
• European gas storage fill — Not yet fetched
• Petroleum production and stocks — Not yet fetched
• Plant, pipeline and terminal geography — Not yet fetched
• AIS-derived vessel activity — Not yet fetched
• Military expenditure (constant US$) — Not yet fetched

Derived series

Where a number is calculated rather than reported, this is the calculation.

Seasonal range bands

The shaded band on a chokepoint or storage chart is the daily minimum and maximum of the previous five complete years, matched by calendar day. It is not a confidence interval and not a forecast: it is the envelope of what that day of the year has looked like before.

Rolling means

Daily transit and event counts are dominated by weekday effects and single-day reporting gaps. Where a chart draws a rolling mean it says so in its subtitle, and the window is centred. The table view under every chart and the CSV download both carry the unsmoothed values.

Change against a baseline

"Vs the 30-day average" compares the latest reading with the mean of the preceding 30 days ending on that reading's own date — not on today's date, so a source that has paused does not manufacture an apparent collapse.

UN General Assembly voting affinity

The affinity figure is the share of votes on which two states cast the same recorded position, per session. Specifically:

• plenary roll-call votes only, and all resolutions — not an important-votes-only subset, which is a choice that changes the answer;
• a pair agrees when both record the same position of Yes, No or Abstain;
• abstentions count as a position, and two abstentions agree;
• absences and non-membership are excluded from both the numerator and the denominator;
• pairs with fewer than 10 shared recorded votes in a session are not reported.

Reasonable people choose differently on the abstention question in particular. The underlying vote records are published at /data so the calculation can be redone on other assumptions.

Leak-site postings

Ransomware leak-site counts are claims made by the groups themselves, not confirmed incidents. They are a measure of what extortion groups assert publicly, which is a real and useful signal, but it is not a victim count and should never be read as one.

How the charts are drawn

• Never two y-axes. Two measures on two scales manufacture correlations that are not there. Two measures get two charts, small multiples, or an index to a common base.
• One palette across all seven sites. Colour follows the series, not its rank, and it does not change when a chart is cross-posted to another domain. The palette is checked computationally for colour-blind separation rather than by eye.
• Identity is never colour alone. Every chart with two or more series carries a legend, and every chart carries a table view of its own numbers.
• Partial periods are marked. The current month or quarter is almost always incomplete; showing it unmarked would read as a collapse.
• Titles state the finding where there is one, rather than naming the variable.

Confidence labels

Analytical pieces can carry a confidence label. It describes how much weight the argument can bear, not how strongly the author feels.

High

High — the claim rests on published data or on-record sources.

Moderate

Moderate — the reasoning is sound but a key input is contested or partial.

Low

Low — a defensible reading of thin evidence.

Speculative

Speculative — an explicit hypothesis, published to be argued with.

Machine-written copy

Some material on these sites is written by a language model. Where it is, it is published under a byline that is not a person's name — currently Wedgetail Desk — and the page says so on its face. We do not put machine-written copy under a human byline.

This currently covers the weekly briefing: the indicators it leads with are selected mechanically from the data layer by threshold, the prose is generated from that selection and the week's published work, and an editor reviews every issue before it goes out. The point of generating it rather than writing it is consistency of selection — the same threshold applies every week whether or not the result is interesting.

Every figure a briefing quotes links to the dataset it came from, so nothing rests on trusting the generated text.

Corrections

Corrections are logged with a date and a description at /corrections, and they also appear on the piece they apply to, so a reader arriving from a link sees the correction without having to go looking.

Licences

Each dataset carries its source's own licence, shown under every chart drawn from it and listed at /data. Attribution requirements travel with the data: they are in the JSON, in the CSV header, and in the chart's source line.

Derivative-note: charts built on V-Dem indicators inherit V-Dem's Creative Commons ShareAlike clause — a derived chart carries compatible terms rather than ours alone. Each such chart's source line names V-Dem; reusers of the derived numbers should check V-Dem's own terms alongside ours.